Github user brkyvz commented on a diff in the pull request:
https://github.com/apache/spark/pull/15235#discussion_r80582702
--- Diff:
sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/ListingFileCatalog.scala
---
@@ -82,73 +85,185 @@ class ListingFileCatalog(
* This is publicly visible for testing.
*/
def listLeafFiles(paths: Seq[Path]): mutable.LinkedHashSet[FileStatus] =
{
- if (paths.length >=
sparkSession.sessionState.conf.parallelPartitionDiscoveryThreshold) {
- HadoopFsRelation.listLeafFilesInParallel(paths, hadoopConf,
sparkSession)
- } else {
- // Right now, the number of paths is less than the value of
- // parallelPartitionDiscoveryThreshold. So, we will list file
statues at the driver.
- // If there is any child that has more files than the threshold, we
will use parallel
- // listing.
-
- // Dummy jobconf to get to the pathFilter defined in configuration
- val jobConf = new JobConf(hadoopConf, this.getClass)
- val pathFilter = FileInputFormat.getInputPathFilter(jobConf)
-
- val statuses: Seq[FileStatus] = paths.flatMap { path =>
- val fs = path.getFileSystem(hadoopConf)
- logTrace(s"Listing $path on driver")
-
- val childStatuses = {
- try {
- val stats = fs.listStatus(path)
- if (pathFilter != null) stats.filter(f =>
pathFilter.accept(f.getPath)) else stats
- } catch {
- case _: FileNotFoundException =>
- logWarning(s"The directory $path was not found. Was it
deleted very recently?")
- Array.empty[FileStatus]
- }
- }
+ val files =
+ if (paths.length >=
sparkSession.sessionState.conf.parallelPartitionDiscoveryThreshold) {
+ ListingFileCatalog.listLeafFilesInParallel(paths, hadoopConf,
sparkSession)
+ } else {
+ ListingFileCatalog.listLeafFilesInSerial(paths, hadoopConf)
+ }
+
+ mutable.LinkedHashSet(files: _*)
+ }
+
+ override def equals(other: Any): Boolean = other match {
+ case hdfs: ListingFileCatalog => paths.toSet == hdfs.paths.toSet
+ case _ => false
+ }
+
+ override def hashCode(): Int = paths.toSet.hashCode()
+}
+
+
+object ListingFileCatalog extends Logging {
+
+ // `FileStatus` is Writable but not serializable. What make it worse,
somehow it doesn't play
+ // well with `SerializableWritable`. So there seems to be no way to
serialize a `FileStatus`.
+ // Here we use `SerializableFileStatus` to extract key components of a
`FileStatus` to serialize
+ // it from executor side and reconstruct it on driver side.
+ private case class SerializableBlockLocation(
+ names: Array[String],
+ hosts: Array[String],
+ offset: Long,
+ length: Long)
+
+ private case class SerializableFileStatus(
+ path: String,
+ length: Long,
+ isDir: Boolean,
+ blockReplication: Short,
+ blockSize: Long,
+ modificationTime: Long,
+ accessTime: Long,
+ blockLocations: Array[SerializableBlockLocation])
+
+ /**
+ * List a collection of path recursively.
+ */
+ private def listLeafFilesInSerial(
+ paths: Seq[Path],
+ hadoopConf: Configuration): Seq[FileStatus] = {
+ // Dummy jobconf to get to the pathFilter defined in configuration
+ val jobConf = new JobConf(hadoopConf, this.getClass)
+ val filter = FileInputFormat.getInputPathFilter(jobConf)
+
+ paths.flatMap { path =>
+ logTrace(s"Listing $path")
+ val fs = path.getFileSystem(hadoopConf)
+
+ // [SPARK-17599] Prevent ListingFileCatalog from failing if path
doesn't exist
+ val status: Option[FileStatus] = try Option(fs.getFileStatus(path))
catch {
+ case _: FileNotFoundException =>
+ logWarning(s"The directory $path was not found. Was it deleted
very recently?")
+ None
+ }
- childStatuses.map {
- case f: LocatedFileStatus => f
-
- // NOTE:
- //
- // - Although S3/S3A/S3N file system can be quite slow for
remote file metadata
- // operations, calling `getFileBlockLocations` does no harm
here since these file system
- // implementations don't actually issue RPC for this method.
- //
- // - Here we are calling `getFileBlockLocations` in a sequential
manner, but it should not
- // be a big deal since we always use to
`listLeafFilesInParallel` when the number of
- // paths exceeds threshold.
- case f =>
- if (f.isDirectory ) {
- // If f is a directory, we do not need to call
getFileBlockLocations (SPARK-14959).
- f
- } else {
- HadoopFsRelation.createLocatedFileStatus(f,
fs.getFileBlockLocations(f, 0, f.getLen))
+ status.map(listLeafFiles0(fs, _, filter)).getOrElse(Seq.empty)
+ }
+ }
+
+ /**
+ * List a collection of path recursively in parallel (using Spark
executors).
+ * Each task launched will use [[listLeafFilesInSerial]] to list.
+ */
+ private def listLeafFilesInParallel(
+ paths: Seq[Path],
+ hadoopConf: Configuration,
+ sparkSession: SparkSession): Seq[FileStatus] = {
+ assert(paths.size >=
sparkSession.sessionState.conf.parallelPartitionDiscoveryThreshold)
+ logInfo(s"Listing leaf files and directories in parallel under:
${paths.mkString(", ")}")
+
+ val sparkContext = sparkSession.sparkContext
+ val serializableConfiguration = new
SerializableConfiguration(hadoopConf)
+ val serializedPaths = paths.map(_.toString)
+
+ // Set the number of parallelism to prevent following file listing
from generating many tasks
+ // in case of large #defaultParallelism.
+ val numParallelism = Math.min(paths.size, 10000)
+
+ val statuses = sparkContext
+ .parallelize(serializedPaths, numParallelism)
+ .mapPartitions { paths =>
+ val hadoopConf = serializableConfiguration.value
+ listLeafFilesInSerial(paths.map(new Path(_)).toSeq,
hadoopConf).iterator
+ }.map { status =>
+ // Turn FileStatus into SerializableFileStatus so we can send it
back to the driver
+ val blockLocations = status match {
+ case f: LocatedFileStatus =>
+ f.getBlockLocations.map { loc =>
+ SerializableBlockLocation(
+ loc.getNames,
+ loc.getHosts,
+ loc.getOffset,
+ loc.getLength)
}
+
+ case _ =>
+ Array.empty[SerializableBlockLocation]
}
- }.filterNot { status =>
- val name = status.getPath.getName
- HadoopFsRelation.shouldFilterOut(name)
- }
- val (dirs, files) = statuses.partition(_.isDirectory)
+ SerializableFileStatus(
+ status.getPath.toString,
+ status.getLen,
+ status.isDirectory,
+ status.getReplication,
+ status.getBlockSize,
+ status.getModificationTime,
+ status.getAccessTime,
+ blockLocations)
+ }.collect()
- // It uses [[LinkedHashSet]] since the order of files can affect the
results. (SPARK-11500)
- if (dirs.isEmpty) {
- mutable.LinkedHashSet(files: _*)
- } else {
- mutable.LinkedHashSet(files: _*) ++
listLeafFiles(dirs.map(_.getPath))
+ // Turn SerializableFileStatus back to Status
+ statuses.map { f =>
+ val blockLocations = f.blockLocations.map { loc =>
+ new BlockLocation(loc.names, loc.hosts, loc.offset, loc.length)
}
+ new LocatedFileStatus(
+ new FileStatus(
+ f.length, f.isDir, f.blockReplication, f.blockSize,
f.modificationTime, new Path(f.path)),
+ blockLocations)
}
}
- override def equals(other: Any): Boolean = other match {
- case hdfs: ListingFileCatalog => paths.toSet == hdfs.paths.toSet
- case _ => false
+ /**
+ * List a single path, provided as a FileStatus, in serial.
+ */
+ private def listLeafFiles0(
+ fs: FileSystem, status: FileStatus, filter: PathFilter):
Seq[FileStatus] = {
+ logTrace(s"Listing ${status.getPath}")
+ val name = status.getPath.getName.toLowerCase
+ if (shouldFilterOut(name)) {
+ Seq.empty[FileStatus]
+ } else {
+ val statuses = {
+ val (dirs, files) =
fs.listStatus(status.getPath).partition(_.isDirectory)
+ val stats = files ++ dirs.flatMap(dir => listLeafFiles0(fs, dir,
filter))
--- End diff --
I think the directories should be submittable as a parallel job if we were
told that we should parallelize file listing.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
